VANE / CREATIVE FIELD GUIDE
From character creation to your finished AI UGC video. Edition 1.0.
A vertical, approximately 15-second product video: a believable fictional creator, a clear product moment, a short spoken message, and a useful call to action. Finish with a clean exported video and three hook variations. This is an instructional kit, not a set of generated videos or a Higgsfield subscription.
Brief > character image > product references > opening image > script > Seedance 2.5 generation > review > edit > export. Character creation uses an image tool inside Higgsfield, such as Soul 2.0; Seedance 2.5 is the video stage, not the character-image model.
Sign in at higgsfield.ai. Check that Seedance 2.5 is available in your account, the generation settings offered, and the displayed cost. If the model is absent, stop and confirm availability rather than choosing another model under the same instructions. Access, menus and prices can change.
Read the guide once, then work through one stage at a time. Copy the prompts from the companion TXT file and replace all bracketed fields. Use the brief and review checklist before making more variations. The fictional example is a reusable coffee cup called DAY CUP; replace it with your own product facts.
Verified feature notes cite sources [1]-[4] on the last page. Prompts, timings, editing suggestions and checklists are VANE recommendations, not official presets or guaranteed results. Researched 16 October 2026. No paid account generation or export was performed to test this kit.
Record your product name, audience, the situation they use it in, one verified feature, the desired action, and the landing-page URL. Add your approved wording and anything the video must not claim. Keep the first version focused on one audience and one message.
Create folders called 01-brief, 02-character, 03-product, 04-stills, 05-generations and 06-final. Keep original files separate from generated versions. Use simple names such as creator-master-v1.png and product-front-original.jpg.
Use a reference to study the opening, rhythm and framing. Write down what you like in your own words. Create your own sequence rather than reproducing a creator’s exact identity, testimonial or full script. Use your own product images and assets you can use.
Product: DAY CUP, a fictional reusable coffee cup. Audience: people making a morning coffee at home. Approved visual facts: a cream body, dark green lid and a reusable form. Message: a cup for an everyday coffee routine. CTA: Explore DAY CUP. We do not claim heat retention, leakproof performance or environmental savings.
You can describe the concept in one sentence, your product photos exist, and every claim has a source. Save the completed brief before generating the character.
In Higgsfield, open Image and choose Soul 2.0 if it is available. Generate an original adult character from a text description. Higgsfield documents using a Soul-generated character as an avatar and using Soul ID to help reuse identity [2,3]. You can instead use a permitted real-person reference.
Create a photorealistic vertical portrait of a fictional adult creator, approximately 30 years old, with shoulder-length dark wavy hair, brown eyes and natural skin texture. Plain oatmeal T-shirt, soft window light, neutral kitchen background, chest-up composition at phone-camera height. Relaxed expression, both eyes sharp, natural proportions. No logos, text, beauty filter or resemblance to a public figure. This is the master identity reference for a product video.
Compare a small batch rather than endlessly changing the description. Inspect the eyes, teeth, ears, hairline, neck and visible fingers. Choose one image with an unobstructed face. A good-looking image with obvious anatomical mistakes is not a good master reference.
Download the selected image as creator-master-v1. Record hair, clothing, accessories and lighting in the brief. Reuse this image in later generations. If you use Soul ID, follow its current creation instructions; do not assume a single-photo reference and a trained identity are the same feature [3].
For a first video, use one adult character and a simple outfit. A second person, intricate jewellery or frequent wardrobe changes creates more details to review. Do not generate a new face for every scene.
Take clear front, side and three-quarter photos in even light. Include an unobstructed packaging reference. Higgsfield’s product workflow supports a saved product and multiple views; it also offers avatar-led product images [2]. Product-reference slots and video-reference slots may live in different workspaces.
Write its exact silhouette, colour, cap or lid, material, label position and approximate scale. For DAY CUP: cream cylindrical cup, dark green removable lid, simple unbranded body. Keep this description unchanged when you test different hooks.
Use your character reference and product photo in a reference-capable image workflow available in your account. Request a chest-up image with the character holding the product beside the chest, face unobstructed. If your chosen image tool cannot accept both references, switch to a supported image workflow rather than relying on text alone.
Use the attached creator as the same person and the attached product as the exact item. Show the creator in a softly lit kitchen holding the product upright beside their chest, label facing camera, with a simple natural grip. Keep the reference face, clothing, product proportions and colours. Vertical phone-photo composition with space above the head. No extra items or generated captions.
Compare the still with the original photos. Reject changing lids, misspelled labels, duplicated fingers and incorrect scale. Use the approved still as the video’s visual starting point where the interface supports it. Add small print and precise branding in the editor when generation cannot reproduce them reliably.
VANE starting structure: 0-3s hook; 3-8s product introduction or demonstration; 8-12s one useful detail; 12-15s CTA and a steady ending. Speak the lines aloud with a timer. If they feel rushed, shorten them before generation. Timing is a creative target, not a guarantee.
0-3s: creator looks into camera: "A small part of my morning coffee setup." 3-8s: gently raises the cup: "This is DAY CUP, a reusable cup for everyday coffee." 8-12s: cut to a simple product close-up; no dialogue. 12-15s: return to creator: "Take a closer look at DAY CUP."
Curiosity: "One small change to the morning coffee setup." Product-first: "Meet DAY CUP, a reusable coffee cup." Routine: "Coffee, a quiet kitchen, and a cup ready to go." These are creative examples, not real customer statements.
Give the character one manageable action at a time: look, lift, hold, set down. Put complicated opening, pouring or assembly in a separate insert if needed. Avoid asking for a pour, a camera spin, a wardrobe change and dialogue in the same moment.
A fictional creator should not claim a personal purchase, weeks of testing or a real result. Write a product introduction rather than a fabricated testimonial. Use only approved offer wording. Decide on any AI-content disclosure needed for your publishing destination before the final edit.
Open Higgsfield’s Seedance 2.5 page [1]. Its published basic flow is to add a reference image, write the prompt, then generate and export. In your signed-in account, confirm the model name and review the controls actually shown.
Add the approved opening still and, where supported, separate creator and product references. Keep a small set with a clear purpose. Identify each attached file by the reference name or token assigned in the UI. Do not paste a made-up @Image token and assume the tool will resolve it.
VANE recommends 9:16, 15 seconds and 1080p when those options are available. Higgsfield’s prompting guide describes up to 30 seconds, native audio and output up to 1080p [4]. Its model landing page includes different 4K marketing language [1], so use the actual generation/export options in your account; do not assume a native resolution.
Copy PROMPT 3 from the TXT file. Replace CREATOR_REFERENCE and PRODUCT_REFERENCE with your interface’s assigned references, or plainly describe which upload serves each role. Replace the example script. Keep dialogue short and describe the ending explicitly.
Read the displayed credit cost before clicking Generate. Save the prompt, settings and output as v01. Watch the whole result with headphones and once without sound. Do not batch ten variations until this first version has a stable face, an accurate product and understandable speech.
Separate visual style, scene, reference roles, action timing and audio. Higgsfield recommends structured prompts and explicit reference roles [4]. The following VANE example is original and should be adapted to your own product.
STYLE: Vertical 9:16, 15-second UGC-style product introduction, natural phone-camera texture and soft window light. CREATOR: Use CREATOR_REFERENCE for the same adult face, hair and oatmeal T-shirt. PRODUCT: Use PRODUCT_REFERENCE for the cream cup and dark green lid. LOCATION: One quiet kitchen. Keep its layout and lighting stable.
0-3s: chest-up, eye-level view; creator says, "A small part of my morning coffee setup." 3-8s: creator gently lifts the cup beside their chest and says, "This is DAY CUP, a reusable cup for everyday coffee." 8-12s: hard cut to a close view of the cup resting on the counter. 12-15s: return to the creator saying, "Take a closer look at DAY CUP." End with a calm visible hold, not black.
AUDIO: Conversational adult voice, clearly spoken English, accurate lip sync, quiet room ambience, no music. Finish speech before the final hold. LIMITS: Same face, wardrobe and product throughout. No invented captions, logos, extra objects or claims. Natural small gestures; do not obscure the face with the cup.
If the product insert breaks continuity, generate a simpler talking-head version and a separate product insert. Assemble them in an editor. A clear two-shot video is more useful than an ambitious sequence with visible errors.
Compare the first, middle and last frames against the master. Reuse the same reference; remove conflicting descriptions or extra face images. Simplify the camera move and return to a single setting. For the next attempt, change one variable and keep the rest of the brief.
Use a sharper reference, a simpler grip and less rotation. Keep the product in view without covering the face. If only the product insert is bad, replace that insert rather than discarding a good creator performance. Use original product photography for a precise final packshot.
Shorten the script, remove overlapping sound instructions and leave time for breathing. If the account provides a targeted edit tool, check its current limits before using it. Alternatively regenerate a shorter speaking shot. Do not conceal inaccurate speech behind subtitles.
Reduce finger motion and object transfers. Request a steady hold or place the product on a counter. Review contact points frame by frame. More prompt adjectives will not necessarily repair a complex action; a simpler shot often gives you a clearer next test.
Specify a visible final hold and no generated text. Trim a bad tail in the editor if the message remains complete. Slow down the blocking and remove unnecessary camera motion. Record the failure and fix in the checklist so the next render tests a specific hypothesis.
Download the approved generated clips. Use CapCut, Premiere, DaVinci Resolve or an editor you know. This editing route is a VANE recommendation, not a claim that Higgsfield has these exact timeline controls. Start a vertical project and place the hook first.
Remove empty opening frames, awkward pauses and unfinished motion at the end. Put the product insert over a clean spoken phrase if it helps clarity. Avoid decorative transitions that distract from the demonstration. Preserve a clear final frame with the product or creator visible.
Create captions from the actual final audio, then proofread every word. Use short phrases, readable contrast and spacing. Keep important text away from where the destination app places its buttons and captions. Preview at phone size rather than trusting the desktop layout.
Keep dialogue intelligible and avoid clipping. If you add music, use a track you can use and keep it below speech. Listen on headphones and a phone speaker. A quiet-room sound bed can be sufficient; a music track is optional.
Add a short CTA matching the destination page, and only an approved offer. Use the real product photo if generated packaging text is unstable. Include an AI-created label where appropriate for your use. Keep an editable master so you can update links or offer wording later.
For a common vertical delivery, start with an MP4 using H.264 at 1080 x 1920 if your source supports it, matching the project frame rate. This is a general VANE delivery recommendation, not a Seedance setting. Upscaling a lower-resolution clip does not create the original detail of a higher-resolution render.
Use the current placement requirements for the platform receiving the video. Confirm accepted duration, size, format, music rights and disclosure options there. Do not rely on a single fixed export preset for every channel.
Open the actual exported MP4, not just the timeline. Check for black frames, missing audio, clipped words, soft text, an incorrect crop and a broken final CTA. Scrub the hands, product and face at the start, middle and end.
Use a name such as daycup_hook-curiosity_v01_9x16.mp4. Deliver the final MP4, a thumbnail, the script, the approved references and the editable project if required. Keep those assets together with the brief and the model/settings record.
Make three variants by changing only the hook first. Keep product, offer, audience and body script constant. Record the version, change, placement and result. A generated video is not evidence that an ad will perform; use your actual campaign results to choose the next test.
[ ] Product and audience defined. [ ] Approved facts and CTA written. [ ] Product photos checked. [ ] Character master selected. [ ] Opening still approved. [ ] Script read aloud with a timer. [ ] Correct model and displayed cost checked.
[ ] Same identity throughout. [ ] Product matches source photos. [ ] Hands and contact points look plausible. [ ] Voice and mouth movements agree. [ ] All lines finish. [ ] No invented text or claims. [ ] Final frame is visible.
[ ] Captions proofread. [ ] Music and reference permissions checked. [ ] AI-content disclosure reviewed. [ ] Actual exported file watched. [ ] Destination format checked. [ ] CTA destination works. [ ] Prompt, references and editable master saved.
If the multi-shot version keeps failing, reduce scope: one creator, one setting, one product held steadily, one short product-introduction line and one CTA. Use an authentic product photograph as the final image. Build complexity only after this version works.
Duplicate the successful prompt, change the first sentence and opening action, and save it as a new version. Do not overwrite the approved version. Write one sentence explaining what you expect this variant to teach you.
[1] Higgsfield: Seedance 2.5 model page. https://higgsfield.ai/seedance/2.5
[2] Higgsfield: How to Make AI Product Photos Without a Studio. https://higgsfield.ai/blog/how-to-make-ai-product-photos-without-a-studio
[3] Higgsfield: How to Create and Use a Soul ID Character. https://higgsfield.ai/creator-hub/help-center/ai-models/how-do-i-create-and-use-a-soul-id-character
[4] Higgsfield: Seedance 2.5 Complete Prompting Guide. https://higgsfield.ai/blog/seedance-2-5-prompting-guide
VANE-Seedance-2.5-Guide.pdf: this guide.
VANE-Prompts.txt: copyable original prompts.
VANE-Campaign-Brief.md: reusable planning template.
VANE-Review-Checklist.md: review and handoff checks.
README.txt: start order and scope.
Edition 1.0 / 16 October 2026. This independent VANE guide is not produced or endorsed by Higgsfield or ByteDance. It documents a suggested creator workflow using published features, not a promise of perfect identity, packaging, audio, performance or generation speed. Verify account controls and current availability when you use it.
If a menu changes, find the same function in the current workspace and use its help documentation. Keep the character, product and script process even when a button moves. Record the date and model version used with each project.